3 ms·
This raises an interesting point. Do you need to train models using CSAM so that the model can self-enforce restrictions on CSAM? If so, I wonder what moral/eth
by giantg2 11mo ago
This raises an interesting point. Do you need to train models using CSAM so that the model can self-enforce restrictions on CSAM? If so, I wonder what moral/ethical questions this brings up.
- jsheard 11mo agoIt's a delicate subject but not an unprecedented one. Automatic detection of already known CSAM images (as opposed to heuristic detection of unknown images) has been around for much longer than AI, and for that service to exist someone has to handle the actual CSAM before it's reduced to a perceptual hash in a database. Maybe AI-based heuristic detection is more ethically/legally fraught since you'd have to stockpile CSAM to train on, rather than hashing then destroying your copy immediately after obtaining it.
- tcfhgj 11mo ago> Maybe AI detection is more ethically fraught since you'd need to keep hold of the CSAM until the next training run, why? the damage is already done
- tremon 11mo agoWhy would you think that? Every distribution, every view is adding damage, even if the original victim doesn't know (or even would rather not know) about it.
- jjk166 11mo agoI don't think AI training on a dataset counts as a view in this context. The concern is predators getting off on what they've done, not developing tools to stop them.
- pseudalopex 11mo agoDebating what counts as a view is irrelevant. Some child pornography subjects feel violated by any storage or use of their images. Government officials store and use them regardless.
- tcfhgj 11mo agoI don't think it's how it works.
- pseudalopex 11mo agoSome victims feel this way. Some do not.
- giantg2 11mo agoI would think there's more possibility of it leaking or being abused in a giant stockpile. Undoubtedly, those training sets would be commercialized in some way, potentially causing some to see that as adding insult to injury.
- boothby 11mo agoI know what porn looks like. I know what children look like. I do not need to be shown child porn in order to recognize it if I saw it. I don't think there's an ethical dilemma here; there is no need if LLMs have the capabilities we're told to expect.
- cs02rm0 11mo agoThey don't have your capabilities.
- mossTechnician 11mo agoI've seen AI image generation models described as being able to combine multiple subjects into a novel (or novel enough) output e.g. "pineapple" and "skateboarding" becomes an image of a skateboarding pineapple. It doesn't seem like a reach to assume it can do what GP suggests.
- jjk166 11mo agoAI doesn't know what either porn or children are. It finds correlations between aspects of inputs and the labels porn and children. Even if you did develop an advanced enough AI that could develop a good enough idea of what porn and children are, how would you ever verify that it is indeed capable of recognizing child porn without plugging in samples for it to flag?
- wang_li 11mo agoSo it is able to correlate an image as porn and also correlate an image as containing children. Seems like it should be able to apply an AND operation to this result and identify new images that are not part of the data set.
- Nevermark 11mo agoThe AI doesn’t even need to apply the AND. Two AI queries. Then AND the results with one non-AI operation.